Towards a Multidimensional Poverty Index for Germany

A Service of

Leibniz-Informationszentrum econstor Wirtschaft Leibniz Information Centre Make Your Publications Visible. zbw for Economics

Suppa, Nicolai

Conference Paper Towards a Multidimensional Poverty Index for Germany

Beiträge zur Jahrestagung des Vereins für Socialpolitik 2015: Ökonomische Entwicklung - Theorie und Politik - Session: Public Choice and Welfare, No. G12-V2

Provided in Cooperation with: Verein für Socialpolitik / German Economic Association

Suggested Citation: Suppa, Nicolai (2015) : Towards a Multidimensional Poverty Index for Germany, Beiträge zur Jahrestagung des Vereins für Socialpolitik 2015: Ökonomische Entwicklung - Theorie und Politik - Session: Public Choice and Welfare, No. G12-V2, ZBW - Deutsche Zentralbibliothek für Wirtschaftswissenschaften, Leibniz-Informationszentrum Wirtschaft

This Version is available at: http://hdl.handle.net/10419/113086

Standard-Nutzungsbedingungen: Terms of use:

Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes.

Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise use the documents in public. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated licence. www.econstor.eu Towards a Multidimensional Poverty Index for Germany

Nicolai Suppa∗ August 2015

Abstract This paper compiles a multidimensional poverty index for Germany. Drawing on the capability approach as conceptual framework, I apply the Alkire-Foster method using German data. I propose a comprehensive operationalization of a multidimensional poverty index for an advanced economy like Germany, including a justification for several dimensions. Income, however, is rejected as a dimension on both conceptual and empirical grounds. I document that insights obtained by the proposed multidimensional poverty index are consistent with earlier findings. Moreover, I exploit the decomposability of the Alkire-Foster measure for both a consistently detection of specific patterns in multidimensional poverty and the identification of driving factors behind its changes. Fi- nally, the results suggest that using genuine multidimensional measures makes a difference. Neither a single indicator nor a dashboard seem capable of replacing a multidimensional poverty index. Moreover, I find multidimensional and income-poverty measures to disagree on who is poor.

Keywords: multidimensional poverty, Alkire-Foster method, capability approach, SOEP

JEL Classiﬁcation Numbers: I3, I32, D63, H1

∗TU Dortmund, Department of Economics, 44221 Dortmund, Germany, e-mail: nicolai.suppa@tu- dortmund.de, phone: +49 231 755-4374, fax: +49 231 755-5404. The author gratefully acknowledges funding by the German Research Foundation (DFG).

1 1 Introduction

Background. The last two decades have witnessed increasing interest in both concepts and measures of well-being. Remarkable efforts have been made, from the Human De- velopment Index in 1990, to the Millennium Development Goals in 2001, to the OECD Better Life Index in 2011.1 Conceptual frameworks related to well-being, such as the capability approach (CA), the subjective well-being literature, and the theory of fairness, are burgeoning alike. In 2009 the so-called Stiglitz-Sen-Fitoussi Commission, appointed to explore alternative measures of welfare and social progress, presented its report. By now, the importance of measuring well-being in general and poverty and social exclusion in particular is acknowledged even in advanced economies. Along with these developments, significant improvements in the methodology of multidimensional measurements have been made as well (e.g., Tsui, 2002, Bourguignon and Chakravarty, 2003, Alkire and Foster, 2011a). So far, these measures have been systematically employed to analyze poverty in the developing world; see in particular Alkire and Santos(2011) and UNDP(2011). However, applying these techniques to advanced economies requires appropriately adapted specifications and operationalizations, such as choosing the relevant dimensions, appropriate indicators, and reasonable cutoffs. More- over, these choices are also contingent upon the concrete purpose of the poverty measure: Is the task to identify general trends across countries and to assess countries’ relative performance in fighting poverty? Or alternatively, is there a need for a society-tailored poverty index to evaluate policy measures more carefully and to better understand both poverty structure and dynamics in that society? As these overall objectives crucially affect the response to many of the arising trade-offs, their explication is imperative.

Previous Research. Recent attempts applying the Alkire-Foster method (AFM) to advanced economies include Whelan et al. (2014) and Alkire et al. (2014). Both studies focus on cross-country comparisons and use EU-SILC data, where most indicators are located in resource space. While Whelan et al. (2014) only exploit the cross-section, Alkire et al. (2014, p. 3) emphasize that currently their contribution is not an empirical one, for reasons of data availability and coverage. Busch and Peichl(2010) also apply the AFM (among other methods), using SOEP data. However, they only consider education, health, and income and only loosely relate their work to a conceptual framework. Also using SOEP data, Rippin(2012) employs a different method (a correlation-sensitive poverty in-

1See UNDP(1990),UN(2012), OECD(2011).

2 dex), which also reflects inequality among the poor. However, Alkire and Foster(2013) demonstrate that no measure can be both sensitive to inequality (understood as dimensional transfer) and satisfy dimensional breakdown and subgroup decomposability simultaneously. Moreover, if at hand, most studies include income as a dimension, although it is unclear whether such an approach is justified (conceptually and empirically). Finally, there is also the literature on material deprivation in the tradition of Townsend (1979) and Yitzhaki(1979), thanks to which new indicators have been widely introduced. This research, however, primarily relies on resource indicators. Consequently, their trans- formation into well-being is mostly ignored. Thus, despite some attempts in this direction, more comprehensive and well-justified multidimensional poverty indexes for advanced economies are still lacking.

Contribution. The present paper complements the previous literature in several ways. Conceptually, I propose a more comprehensive operationalization of a MPI for an advanced economy like Germany, including a justification for selected dimensions. Specifically, I argue in favour of including material deprivation and employment as important dimensions, as they contribute extra information on otherwise ignored functionings. However, I reject a lack-of-income dimension on both conceptual and empirical grounds. In addition to education, health, housing, I also propose an operationalization of social participation. Empiri- cally, I demonstrate that insights obtained by the proposed multidimensional-poverty index are consistent with earlier findings (e.g., migrants suffer more poverty). Going beyond a documentation of changes in multidimensional-poverty, I exploit, moreover, features of the adopted method (e.g., its decomposability) that allow to consistently detect specific patterns (e.g., changing gaps or other asymmetric impacts). Unfolding the summary measure allows, moreover, to identify the driving factors behind changes in poverty (e.g., changes employment or material deprivation indicators). Finally, I demonstrate that using genuine multidimensional measures makes a difference. First, the data at hand suggest that neither a single indicator, nor a dashboard approach can replace a genuine multidimensional approach. The crucial information of coupled deprivation (the “joint distribution”) is otherwise easily missed. Importantly, I also find multidimensional- and income-poverty measures to substantially disagree on who is poor. This contrast in targeting renders different policy implications likely.

Signiﬁcance. The present study enhances multidimensional poverty measurement for an advanced economy like Germany. Since, by now, the importance of poverty in advanced

3 economies is widely acknowledged, several governments, started to compile dedicated reports, documenting numerous poverty-relevant developments. The German government, for instance, now releases an official report on poverty and wealth (RPW) for each legisla- tive session. The reports publish and analyze selected core indicators, and also provide advice on policy measures. So far, however, the RPWs lack both a composite measure and a systematic account of multiple deprivation.2 The present study aims to close this gap and promote a multidimensional poverty index tailored to the German society. Such an index complements the official reports with (i) a comprehensive summary measure (which still allows a detailed analysis) that (ii) takes account of the joint distribution of deprivations and (iii) improves the measurement of poverty as capability deprivation. Indeed, the latest RPW finds difficulties in measuring functionings, capabilities, and capability deprivations (see Bundesregierung, 2013, pp. 23–24).

Procedure. A cogent poverty measure must (i) be embedded within a grounded conceptual framework, (ii) have a sound technical basis, and (iii) use high-quality data for the calculation. To meet these requirements I first adopt the CA, essentially as developed by Sen(1985, 1992, 1999b), as a conceptual foundation. Dimensions are understood as functionings, which in turn constitute human well-being. Because of this inherently multidimensional concept of well-being, the CA offers a comprehensive and coherent account of deprivations. Moreover, for the inevitable value judgments (normative exercises) the CA requires any application to draw on a relevant public debate (e.g., Sen, 1999b, ch.6). Second, I apply the dual cutoff counting approach suggested by Alkire and Foster(2011a). The AFM fulfills several desirable axioms that allow a sensible analysis (e.g., numerous decompositions). Moreover, the AFM is sensitive to changes in both the breadth and the incidence of poverty. Importantly, as an “open-source technology”, it also reveals rather than buries the value judgments and thereby allows for a constructive exchange with the public debate. Finally, I use the SOEP, a rich, high-quality data set for Germany, which allows a comprehensive specification.

Outline. Section2 provides a brief exposition of the underlying methods; section3 intro- duces both data and speciﬁcation. Section4 presents the results. Finally, section5 offers some concluding remarks.

2The importance of multiple deprivation or “the joint distribution” has been emphasized repeatedly (Duclos et al., 2006, Wolff and de-Shalit, 2007, Stiglitz et al., 2009, Ferreira, 2011).

4 2 Methodology

The Alkire-Foster method offers numerous beneﬁts for the evaluation of both poverty- relevant developments and policy measures. The exposition here is restricted to those aspects used in the subsequent empirical analysis. Further aspects are found, e.g., in Alkire and Foster(2011a,b). Alkire et al. (2015) provide a more comprehensive discussion.

Identification. The matrix y contains the available data, is of size N D, and describes × for each individual the achievement in each dimension deemed relevant. Specifically, yid ≥ 0 represents the achievement of individual i = 1, . . . , N in dimension d = 1, . . . , D. The row vector z, with zd > 0, describes the deprivation cutoffs, i.e., the achievements necessary for not being considered as deprived in the respective dimension. Using this information, we obtain the deprivation vector c by counting individual deprivations, i.e., the column D vector’s elements are c P 1 y < z . Following Bourguignon and Chakravarty i = d=1 ( id d ) (2003), the discrimination between poor and non-poor individuals depends critically on dimensional achievements and the respective cutoffs. Thus identification can be described by a function ρ(yi, z) . Several approaches have been suggested so far. While the union 1 approach is characterized by ρ(yi, z) = (ci 1), the intersection approach requires ci = ≥ 1 D. The key idea of Alkire and Foster(2011a) is to define ρk(yi, z) = (ci k) for k = ≥ 1, . . . , D. Since ρk depends on both the dimension-specific cutoffs zj and the overall cutoff k, it is called the dual cutoff approach. The union and intersection approaches are included as special cases (k = 1 and k = D).

Aggregation. A simple form of aggregation is the calculation of the headcount ratio, N which is defined as H q/N, where q P 1 c > k is the number of the poor. Addition- = = i=1 ( i ) ally, to take account of the breadth of poverty we first censor the counting vector of depriva- 1 tions for non-poor and thus define c(k) with elements ci(k) = (ci k)ci for all i = 1, . . . , N. N As c k /D is the share of all possible deprivation suffered by i, A≥ P c k / qD repre- i( ) = i=1 i( ) ( ) sents the average deprivation suffered by the poor. Alkire and Foster(2011a) then define the adjusted headcount ratio as M 1 PN c HA, which is sensitive to both changes 0 = N i=1 i = in incidence and breadth of poverty. In principle other members of the FGT class of measures (see Foster, Greer, and Thorbecke, 1984) can be applied as well—their discussion is however beyond the scope of this paper.

Weights. So far we have assumed equal weights for all dimensions. To allow for different weights, we introduce a weighting vector w with PD w 1. Then the weighted d=1 d =

5 PD 1 D PN deprivation count becomes ci = D 1 wd (yid zd ), and M0 = N i 1 ci(k). = ≤ =

Decompositions. The adjusted headcount M0 and both its single components and its changes over time have been shown to be decomposable in numerous ways. For instance, subgroup decomposition for the adjusted headcount ratio means that, after allowing for relative population sizes, the subpopulation-speciﬁc adjusted headcount ratios exactly add up to the overall adjusted headcount ratio. Let the subscript g = 1, . . . , G denote the partic- PG Ng ular subpopulation with g Ng = N and ψg = N . Formally, the subgroup decompositions for the adjusted and the censored headcount ratio then are

G G X X M0(y; z) = ψg M0(yg ; z) and H(y; z) = ψg H(yg , z). (1) g=1 g=1

If data on more than one point of time is available, we also can calculate and decompose changes of aggregate measures. Let the superscript t denote the respective period. The relative change of M0 from t 1 to t then is − M y t ; z M y t 1; z δM t 0( ) 0( − ). (2) 0 M −y t 1; z ≡ 0( − )

The percentage changes of H(k) and A(k) can be deﬁned analogously; in general, however, they are not independent of each other. Consequently, a basic decomposition of the change 3 in M0 is as follows: t t t t t δM0 = δH + δA + δH δA . (3) × Changes in the censored headcount, in turn, can be traced back to subpopulation-speciﬁc headcount ratios, H(yg ; z), and changing shares of the respective subpopulations (ψg ). Formally, G t X t 1 t t t t δH = rg− δψg + δH(yg ; z) + δψg δH(yg ; z) (4) g=1 ×

ψt 1H y t 1;z t 1 g− g ( g− ) with rg− = H y t 1;z being the contribution of subpopulation g to the overall headcount ( − ) in t 1. The adjusted headcount can also be decomposed into the contributions of each dimension− (dimensional breakdown). First, the dimension-speciﬁc censored headcount is

3For an alternative decomposition see Roche(2013), for a comparison along with a discussion of the assumptions, see Alkire et al. (2015, ch. 9.2).

6 1 PN 1 H d N i 1 (ci k yid zd ) , allowing us to rewrite the adjusted headcount as ≡ = ≥ ∧ ≤ D X w M d H . (5) 0 = D d d=1

H Then, the contribution of dimension d to overall poverty is wd d . Additionally, changes in D M0 the adjusted headcount can be decomposed into changes in dimension-speciﬁc censored headcount ratios. Speciﬁcally, D t X t 1 δM0 = sd− δH d , (6) d=1 t 1 t 1 θd Ad (y − ;z) where sd− = A y t 1;z is the contribution of dimension d to the average intensity. ( − )

The Alkire-Foster method and Capability Deprivation. The latest RPW finds difficulties in measuring functionings, capabilities, and capability deprivations (Bundesregierung, 2013, pp. 23–24).4 Admittedly, functionings are often difficult to measure, but capability deprivation even more so. Either we can assume deprivation for low achievements from the outset, which often still may be justified (e.g., Robeyns, 2005, p. 101). Alternatively, we base this assumption—that functionings not chosen were infeasible—on further information. Using the AFM allows exactly this, since being poor (i.e. capability deprived) requires the simultaneous presence of several low-functioning achievements, thereby lending support to the assumption of an enforced low achievement.5 Hence, exploiting the joint distribution in the identification step of poverty analysis helps to distinguish between (deliberately chosen) low-functioning achievements and (enforced) capability deprivations.

3 Data and Speciﬁcation

Sample. For the analysis I use data of the German Socio-Economic Panel (SOEP) and calculate a multidimensional poverty index for three periods of time (2001–02, 2006–07, 2011–12).6 The SOEP not only allows one observe the same individuals in different years,

4Note that Suppa(2014) argues that even if functionings are difﬁcult to measure and often only imperfect data is at hand, the CA’s conceptual structure is still helpful for revealing the underlying assumptions. 5 Alkire et al. (2015, ch. 6.1) provide a more detailed account. Indeed, M0 can be shown to be a measure of unfreedom in the sense of Pattanaik and Xu(1990), who axiomatically the study measurement opportunity sets. 6I use SOEP data v29.1, provided by the DIW; see Wagner et al. (2007) for more details. The data used in this paper was extracted using the add-on package PanelWhiz for Stata. PanelWhiz (http://www.panelwhiz.eu) was written by Dr. John P.Haisken-DeNew ([email protected]). See Hahn

7 but also provides information on various aspects of a respondent’s life. However, to avoid an overload of the respondents, some questions are only asked every other year (or less frequently), whereas other items are only collected in between these years. Consequently, a comprehensive poverty index can only be calculated for selected years. Moreover, for using the best-suited items simultaneously, I merge two consecutive years into one period. Naturally, this comes at the cost of losing those observations not observed in both years of a period. The natural target population for a study on multidimensional poverty in Germany are the adults living in Germany in the respective year. Consequently, I treat the SOEP as repeated cross-sectional data.7 To account for the complex survey design of the SOEP, the subsequent analyses use sampling weights, which are basically the inverse sampling probabilities (see Goebel et al., 2008).

Operationalization. The importance of the conceptual framework for empirical exercises was already outlined and emphasized by Lazarsfeld(1958). The operationalization of the present study draws on both the capability approach and the German government’s ofﬁcial reports on poverty and wealth. Relying simultaneously on both is possible, since the ofﬁcial reports by now explicitly use the CA as well.8 The capability view not only considers human well-being as inherently multidimensional; moreover, it assigns intrinsic importance to functionings, i.e. the doings and beings individuals have reason to value. Note that intrinsic importance naturally leaves room for instrumental importance as well, as being able to read and write or being healthy illustrate.9 Poverty, then, is understood as capability deprivation, implying both a shortfall in one or several of the functionings deemed relevant and their infeasibility for the individual in question.10 Consequently, indicators of deprivation both (i) need to be located in the functioning space and (ii) need to take account of the functioning’s infeasibility. Moreover, the CA requires value judgments to be exposed rather than concealed, and in

and Haisken-DeNew(2013) and Haisken-DeNew and Hahn(2010) for details. The PanelWhiz-generated DO ﬁle to retrieve the data used here is available from me upon request. Any data or computational errors in this paper are my own. 7Exploiting the panel setup of the data, implies a different concept of the samples’ underlying population, i.e. the individuals living in Germany during the complete period investigated. Hence, such a setup ignores several groups by construction including migrants, individuals who become 18, die or otherwise leave the SOEP during the period investigated. Suppa(2015a) exploits the panel setup of the data. 8Moreover, the RPWs also use another framework, the condition-of-life approach, essentially developed by (Neurath, 1917 [2006], 1937 [2006]). For a comparison of the two approaches see Leßmann(2009). 9For the distinction between intrinsic and instrumental relevance see, e.g., Sen(1999b). 10On poverty as capability deprivation, see in particular Sen(1992, ch. 7) and Sen(1999b, ch. 4).

8 addition they must be subjected to public debate. Only with clear presentation of the normative problem can a public debate about these issues be expected to fulfill its constructive role; see Sen(1999a, p. 10). Value judgments are needed for (i) the selection of functionings included in the index, (ii) the respective deprivation cutoffs, (iii) the assigned weights, and (iv) the poverty cutoff. The official reports provide a first set of indicators, which aim at measuring important functionings. Specifically, so-called core indicators are to be regu- larly reported, and their selection is based on scientific advice (Arndt and Volkert, 2007).11 This selection is thus reasoned and transparent, and yet open to criticism and modification. Hence, the choice of dimensions is subjected to public debate and thereby complies with the aforementioned requirement of the CA (see also Sen, 2004, on this). Subjecting the choice of the deprivation cutoff to public debate, however, further constrains the choice of a functioning’s indicators. Specifically, indicators should allow for deprivation cutoffs that are similar and meaningful across individuals, such that a public debate can study the pros and cons and eventually agree upon those cutoffs. Limitations of available data, however, prompt us to draw on imperfect indicators as well. In some cases a functioning may be only captured incompletely; in others, measurement remains within the resources space. Finally, the CA assigns goods, income, and other resources an instrumental role only, howsoever important they may be.

Dimensions. The increasing interest in alternative measures of well-being motivated numerous novel measurement initiatives in various directions. In addition to that, a consensus on relevant dimensions seems to emerge. Table1 provides an (non-exhaustive) overview of dimensions frequently suggested. Note that Nussbaum(2001) approaches the question from a philosophical and conceptual view. In contrast, the other studies (e.g., Stiglitz et al., 2009) survey and organize already available indicators. As mentioned earlier, exercises in measurement necessitate a clear conceptual understanding. In this respect, the CA argues that if we study poverty or well-being, we ideally measure functionings (or capabilities). As it stands, in many cases we still face imperfect indicators. Some of their shortcomings will be discussed later. Nonetheless, table1 reﬂects an agreement on certain dimensions such as education, health, or social participation. However, table1 also reveals two further aspects. On the one hand, for some of more complex dimensions of human well-being, such as the functionings of self-respect, prac-

11Note that other contributions and debates reach similar conclusions, for instance the report of Stiglitz et al. (2009), or the European efforts for social inclusion (e.g., Atkinson et al., 2002, 2004, Marlier and Atkinson, 2010), but also the earlier Scandinavian approach to welfare (e.g., Allardt, 1993, Erikson, 1993).

9 tical reason, or agency, there is no accepted set of indicators sofar. On the other hand, several dimensions (along with available indicators) are frequently proposed which are, from a conceptual point of view, no functiongings. Consequently, they are not dimensions of well-being by themselves. Leading examples are housing, material deprivation, and income. Conceptually, all of them provide resource information. Only in some cases, resource-indicators can be clearly related to a single key functionging, as e.g. housing indicators (even though conversion factors are then ignored). In constrast, many other indicators are likely to affect several functionings and, moreover, in an a priori unclear way (e.g., employment). Thus, they are best considered as multipurpose means.12 The vital point is this: should we incorporate or ignore information provided by material deprivation indicators, income and other resource-based indicators? Ignoring crucial information about the lives the poor experience, poses a seriously flaw of any poverty measure, just as adding redundant information may distort conclusions as well. The present paper proposes to incorporate resource dimensions, if their indicators—argumentatively or evidentialy—contribute important information on otherwise ignored functionings. For instance, the present studies’ specification introduced later, argues that indicators of material deprivation are well-suited to infer a shortfall in both practical reason and economic security. Recent insights from behavioral economics (also discussed later) lend support to this nexus. In contrast, a shortfall in income, is not included, since a key functioning income may help to achieve is already modelled explicitly, i.e. social participation. This argument is reinforced by including material deprivation indicators which tend to better reflect well-being than income. Consequently, adding an income dimension is likely to cause redundancy—given that material deprivation and social participation indicators are already included. Finally table1, also signals further, partly conceptual, difficulties. For instance, is security best considered as a dimension on its own, as suggested by Stiglitz et al. (2009, p.194) who, however, further distinguish personal and economic security. Or alternatively, is it better to model them as risks for the respective functionings (e.g., bodily integrity or health), as advocated by Wolff and de-Shalit(2007)? Likewise, how to account best for employment-related aspects requires still more investigation and debate.

Specification. Although an in-depth discussion of all indicators is beyond the scope of this study, I briefly comment on the selected indicators. The justification of the material deprivation and employment dimensions receive particular attention. Table2 shows the

12Note that even housing indicators may not only affect “shelter” and “privacy” but also, say, health.

10 selected functionings and their indicators, along with their weights. Note that almost all indicators are either already core indicators of or analyzed within the RPW.13

Education. Education is meant to capture not only achievements in reading and writ- ing, but also the abilities to use one’s senses, to imagine, think, and reason (see Nussbaum, 2001). The ﬁrst indicator (dep_educ) switches to deprivation if a respondent failed to complete elementary education or completed elementary education but later failed to obtain a vocational qualiﬁcation. Elementary education refers to the graduation after Germany’s 10 years of compulsory education. Beyond formal education, I also consider the number of books within the household. Members of a household owning less than 10 books are considered deprived (dep_N books). This information proxies both the educational cli- mate within the household and effective literacy.14 However, as a proxy located in the resource space, it suffers the usual limitations (potentially important conversion factors are ignored).

Health. Deprivation in health, which is multidimensional itself, is signalled by three indicators. First, respondents are deemed deprived of bodily integrity if they are partially or severely disabled (dep_disabilit y). Second, I compile a sub-index, which allows for substitutability among several medical conditions. Two out of four health problems must be reported for being deprived. The four health issues are (i) a strong limitation when climbing stairs, (ii) a strong limitation for tiring activities, (iii) physical pain occured always or often during the last 4 weeks, and (iv) the health condition limited always or often socially. Finally, a BMI larger than 30 (dep_obesit y) indicates, according to WHO(2000, p. 242), obesity and thus is medically critical. Note that for these indicators the deprivation cutoffs are similar and meaningful across individuals—avoiding a common drawback of indicators like subjectively assessed health state or health satisfaction.

Housing. Housing indicators are to capture the functionings of being sheltered and en- joying privacy. To measure housing, I resort to resource indicators. Speciﬁcally, I consider a person to be deprived of adequate shelter and privacy if any of bath, kitchen or toilet is missing in her accommodation (dep_hhf acilities) or if the respondent reports that her

13See, e.g., Bundesregierung(2013, 461–491,) or Wissenschaftszentrum Berlin für Sozialforschung (WZB) and Institut für Arbeitsmarkt- und Berufsforschung (IAB)(2013). 14This indicator is used frequently to study the inﬂuence of constructs like “scholarly culture” of the parental household on children’s educational attainments (see, e.g., Evans et al., 2010), and is, moreover, applied by the OECD as well (see, e.g., OECD, 2014).

11 house either “requires major renovation” or is “ready for demolition” (dep_housecond). Finally, I use a simple overcrowding index (dep_overcrowded), which indicates deprivation if there is less than 1 room per person in the household (see Bundesregierung, 2013, p. 243). However, drawing on these resource indicators ignores relevant conversion factors (e.g., the power relations within the family). Moreover, the housing situation may also contribute to healthy living conditions more generally. In addition, it may support self-respect or facilitate social participation.

Social Participation. The measurement of social participation exploits information on the frequency with which certain activities are reported to be performed. These activities represent common forms of social life. Respondents may report at least once a week, at least once a month, less often, or never. Table A.1 contains the exact wording of the questions. While meeting friends or relatives, the social activity par excellence, is of central importance, many other activities also facilitate relatedness and social interaction. To emphasize the importance of meeting one’s friends (for its own sake), I consider a person deprived if she reports to never meet her friends. The remaining items are used to construct an activity index. Speciﬁcally, the activity index considers an individual deprived if she reports never performing six or all of a list of seven activities or, alternatively, never performing ﬁve activities and, additionally, performing one or two activities less often.

Material Deprivation. Inspired by the work of Townsend(1979) and others, previous poverty measures also used indicators for consumption or ownership on selected goods. Conceptually goods, like income, are resources. Notwithstanding, resource indicators may provide extra information. However, material and wealth deprivation are best considered as a shortfall in a multipurpose means. Lacking multipurpose means may affect several distinct functionings simultaneously and, moreover, in an a priori unclear way. This paper proposes to use resource dimensions, if their indicators argumentatively or evidentialy contribute extra information on otherwise ignored functionings. More speciﬁcally, I argue that indicators of material and wealth deprivation are well-suited to infer a shortfall in both practical reason and economic security. Nussbaum(2001) suggests the functioning practical reason, referring to an individuals’ capacity to act and to plan one’s life, including the ability to perform deliberate and reasoned actions.15 In economic choice theory this corresponds to the activity of balancing

15Though related to agency, both concepts are distinct. Agency refers to the ability to set one’s own goals and eventually strive for them, such as whether to devote one’s life to a country’s independence, to opt for an austere and spiritual life style, or to maximize one’s well-being (e.g., Sen, 1992, ch. 4). In

12 costs and benefits. The proposed justification for material deprivation draws on recent research from behavioral economics. Specifically, Mullainathan and Shafir(2013) argue economic conditions to systematically distort decision-making via the so-called scarcity mindset. Important implications are both focus dividend and tunneling.16 The authors conclude (p. 119), “When we focus so intensely on making ends meet now, we plan less effectively for the future.” Later (pp. 120–121), they continue, “myopia is not a personal failure. Tunneling is not a personal trait. [...] rather, it is the context of scarcity that makes us all act that way.”17 Individuals struggling hard to make ends meet are fully occupied with monitoring every penny spent and any penny to be earned. Consequently, long-run effects (costs or benefits) and goals are located outside the tunnel, and hence ignored. Since it is these economic conditions that induce (inter alia) myopia, decision making is systematically distorted. Material and wealth deprivation are also suited to signal a lack of economic security. Goods not consumed for financial reasons already indicate difficulties to make ends meet, and thus a threatened level of consumption. Moreover, the role of wealth (and borrowing) in consumption smoothing is well-established, since the permanent income hypothesis of Friedman(1956 [2015]). Finally, depending on the specific goods used, material deprivation indicators may also indicate shortfalls in even other functionings (e.g., respecting oneself). The dimension of material deprivation is operationalized using two sub-indices, which allows among other things a certain substitutability. First, dep_weal th equals one if none of the following wealth items is owned: life insurance, pension, house or apartment, financial assets, commercial enterprise, tangible assets.18 Second, dep_matdep equals one if two or more items of the following are missing for financial reasons (i) a warm meal, (ii) friends are invited for dinner, (iii) money is put aside for emergencies, and (iv) worn out furniture is replaced. Both indicators are suited to detect shortfalls in practical reason and both indicators capture important aspects of economic insecurity. Consequently, extra information is added which otherwise would have been ignored.

contrast, practical reason refers also to technical and operational decisions. However, low achievements in practical reason may well entail deprivation in agency. 16Poorer people, for instance extract a focus dividend as they are found to be robust to commonly found framing effects (Mullainathan and Shaﬁr, 2013, ch. 4, surveys the evidence). 17Shah et al. (2012), Mani et al. (2013) provide more evidence and elaborate this line of thought. 18The absence of wealth items indicates what Mullainathan and Shaﬁr(2013, ch. 3,) call slack (dep_weal th). In their suitcase-packing metaphor, slack is space accidentally left here and there. Among other things, slack also provides room to fail, i.e., less disastrous consequences of erroneous actions.

13 Employment. Previous studies either include an employment dimension or explicitly advocate an employment capability (e.g., Leßmann and Bonvin, 2011, Alkire et al., 2014). In fact, by now there is widespread agreement about the importance of employment for human well-being (e.g., Stiglitz et al., 2009, Bundesregierung, 2013). Employment, or labour more general, may indeed help to do things which are intrinsically important (e.g., masterly proficiency, or contributing one’s share for the good of all).19 However, an array of effects of labour on other dimensions of well-being has been documented as well. In fact, most information collected for the labour-well-being nexus usually pertains to its instrumental relevance (e.g., occupational diseases and risks for accidents, various security schemes, workers’ participation in various processes, exposure to adverse conditions, etc.). Suppa(2015b) elaborates the link of labour and well-being, and argues labour to be a crucial means for achieving numerous functionings, such as being healthy, agency, self- respect, practical reason, appearing in public without shame, etc. The effects of unemployment on well-being are a case in point. Research on life satisfaction, for instance, documents the importance of non-pecuniary costs of unemployment for subjective well-being and thus demonstrates their importance in principle (Winkelmann and Winkelmann, 1998). Further results find identity utility to be important (Hetschko et al., 2013), which from a capability view may indicate an effect on being ashamed or respecting oneself. Some studies also directly examine the effect of unemployment on specific functioning achievements. Kunze and Suppa(2014), for instance, find unemployment to reduce social participation, whereas Schmitz(2011) finds no effect on health in general. If, however, perfect measures for all relevant functionings were available, there would be no need to rely on an unemployment indicator. In this sense, the justification for an unemployment indicator would lapse. As it stands, however, accepted measures for many of the more complex functionings are lacking and existing ones might be incomplete. Thus, similar to the material deprivation indicators, employment-related indicators may provide important extra information on otherwise ignored functionings as well. The current specification draws on three employment-related indicators. First, if an individual reports to be registered unemployment dep_unemp equals one. As outlined above deprivation in numerous functionings a likely to accompany unemployment. Sec- ond, dep_underemp equals one if a person reports to involuntarily work less than 30 hours a week. This may be associated by shortfalls in similar functionings, although to a lesser extend. Moreover, part-time jobs are often found to provide lower job quality. Restrict-

19See e.g. Csíkszentmihályi(1990) who discusses the ﬂow-aspect of labour.

14 ing deprivation to the involuntary is important, since for many households part-time work may, in fact, be desirable for improving the work-life balance. Finally, dep_precemp equals one for persons who are marginally employed or in temporary employment. Associated deprivations are in social and economic security and practical reason.

Weights. The main speciﬁcation assigns equal weights to each dimension and, within a dimension, equal weights to each indicator. Consequently, most indicators receive a weight of 1/18, whereas education and social participation indicators receive 1/12 each. Finally, note that full deprivation in employment is only achieved by unemployed (weighted with 1/6). Assigning the other two indicators a weight of 1/18 each, implies an improvement for a formerly unemployed, who ﬁnds a precarious part-time job.

Who is poor? Many instances of the subsequent empirical analysis use a poverty cutoff k = 33, implying an individual is considered poor if she suffers at least 33% of the (weighted) maximal possible deprivation. Moreover, to consider people who are actually poor, is a useful exercise to justify a poverty cutoff. For instance: There is a male respondent aged 34, who (i) failed to obtain a vocational qualiﬁcation, (ii) exhibits obesity, (iii) is considered at least partially disabled, (iv) is currently unemployed, and (v) reports critically low social activity. As his (weighted) deprivation count amounts to 0.4 (2/12 + 2/18+1/6), he is considered multi-dimensionally poor. Alternatively, consider a female respondent aged 76, who (i) failed to complete general elementary education, (ii) reports strong problems with climbing stairs and often physical pain occurred during the last 4 weeks, (iii) calls none of the wealth items her own, (iv) never meets friends, and also (v) reports critically low social activity. This woman is also considered poor, since her weighted deprivation count sums to 0.389. Note that although these deprivations, such as education and unemployment, might even be causally related, each of them inherently diminishes the life the person leads, which is ultimately why we count it.

Deprivation Headcounts. Table3 provides ﬁrst information about deprivation indicators. The uncensored deprivation headcount is simply the share of individuals deprived in a given indicator. Uncensored headcounts for the whole population (total) indicate different levels of prevalence for different dimensions. Housing indicators, for instance, vary from 1–5%. Similarly, employment indicators vary from 4–7%, whereas deprivations in wealth or social participation amount to 20% each. The so-called dashboard approach exclusively relies on these headcount ratios along with their changes (i.e. the marginal distributions).

15 However, the uncensored headcount ratio can also be calculated for certain subgroups, e.g., by poverty status. While only 1% of the non-poor is found unemployed, 32% of the multidimensional poor are. Table3 also reveals that the most prevalent deprivations among the poor are material deprivations (67–79%) and social participation (54–62%). Simi- larly, 47% of all poor are deprived in education. Given the counting approach to poverty, higher prevalences for the poor are to be expected. In fact, prevalences are substantially larger for the poor—often by a triple or more. This finding simply mirrors the fact, that the AFM exploits the joint distribution of deprivation already in the identification step of poverty measurement. Put differently, the AFM uses the joint distribution to distinguish more important (i.e. coupled) from less important (i.e. occasional) deprivations. In addition to this, Table3 also shows the censored headcount ratios, i.e. the share of the population who is poor and deprived in the given indicator. Censored heacounts are the key ingredient for dimensional decompositions. While the censored headcount must be smaller or equal the uncensored headcount, it is important to note that none of the censored headcounts is really close to its uncensored headcount. Thus, virtually no indicator directly implies poverty (i.e. multiple deprivation). The final column contains the share of a given deprivation borne by the non-poor; thus the fraction of a deprivation ignored throughout the subsequent analysis. Note, that for most indicators the non-poor account for 50% or more of a deprivation. The only exception is unemployment, where only 20% of the unemployed are non-poor (which results from the higher weight). Table3 also clearly reflects that even rather widespread deprivations (e.g., in health or social participation) are by themselves not only insufficient to render an individual poor. Furthermore, a significant share of these deprivations is deliberately ignored in the subsequent analysis. More importantly, table3 suggests that neither a single indicator, nor a dashboard approach is capable of replacing the multidimensional approach. The latter is in particular supported by the high shares of deprivation borne by the non-poor. To infer from a declining uncensored headcount ratio what happens to the multiply deprived becomes a doubtful exercise for the data at hand.

4 Results

Aggregate Measures. Figure1 depicts the multidimensional poverty measure M0 (the adjusted headcount ratio), the incidence H (the headcount ratio), and the intensity A (the average number of deprivations suffered by the poor)—each for all three periods and for poverty cutoffs k [25, 50]. Figure1 suggests for both M0 and H an increase from period ∈

16 1 to 2 and decrease from period 2 to 3— independent of k. Average intensity seems to be lowest in 2001/02. In order to obtain a more detailed account of multidimensional poverty, figures2 and 3 contain adjusted and headcount ratios, each computed for specific subgroups. Figure2 (a), for instance, documents that individuals with a background of migration exhibit both a larger M0 and a larger H—in all years for all relevant k. Similarly, figure2 (b) suggests both a higher M0 and a higher H for East Germany—this difference is, however, much less pronounced. Finally, figure2 (c) shows that differences according to age groups are not that clear-cut, since most lines are crossing another. Figure3 uncovers further substantive differences in multidimensional poverty among groups, for k [10, 50]. Specifically, persons in single households tend to experience more poverty than∈ individuals in households of couples, regardless of eventual children in the household (figure3 (a)). Figure3 (b) clearly documents the importance of the father’s education on an individual’s deprivation. Three groups appear to be distinguished: First, persons with fathers completely lacking education or where education is unknown are associated with the highest M0. The second group consists of individuals whose father completed Hauptschule or other schools, while the fourth contains those whose fathers completed Realschule and Abitur. Finally, figure3 (c) suggests both a slightly higher M0 and H for women. Differences in average intensity vary less by subgroup (e.g., for age), see figure A.1. However, respondents living in couples or whose father’s education is Realschule or better display a slightly lower intensity on average. In sum, figure2 and3 document that the insights generated by the adjusted headcount ratio are consistent with earlier findings. The systematic discrimination of individuals with migration backgrounds is just as well documented as the influence of the family background on the offspring’s educational achievements (e.g., Bundesregierung, 2008, ch. IX and III.5).

Contributions of Subpopulations. The previous results suggest certain socio-demographic groups to suffer more from multidimensional poverty. Instead, this paragraph reveals the shares these groups contribute to overall multidimensional poverty, i.e., relative population size is incorporated. Speciﬁcally, using (1), M0 and H can be decomposed into contributions of each subpopulation to overall poverty. Setting to k = 33, ﬁgure4, shows such a decomposition for German states, suggesting that the populous states NRW,

17 20 BAV, and BW contribute the lion’s share to overall multidimensional poverty (M0). Fig- ure4 also shows that 32% of the overall multidimensional poverty ( M0) is contributed by people with a background of migration. Note that this share is disproportionate to their population share (19%). Figure2 (a) also reflects this finding. Finally, figure4 clearly un- derlines the importance of the father’s educational background. More specifically, respondents reporting their fathers to have completed Hauptschule alone make up approximately

2/3. Including those individuals reporting their father’s education to be unknown, uncom- pleted, or absent, the share of multidimensional poverty associated with a handicapped education of the father climbs to ca. 85%. Admittedly, the corresponding population share is 69%. 21 Nonetheless, this ﬁnding emphasizes the role of the educational background of the father in multidimensional poverty.

Dimensional Breakdown. In ﬁgure5 multidimensional poverty ( M0) is further decomposed to each indicator’s contribution using (5). The subsequent ﬁgures report both the

wd absolute contribution of a dimension d, D H d , summing to M0, and the relative contri- H bution, wd d , summing to 100%. Figure5 suggests the dimensional contributions to be D M0 stable over time. The major contribution comes from social participation, followed by a material deprivation. Housing indicators add the smallest share (ca. 4%). In order to display different profiles of poverty, figure6 (a) shows dimensional breakdowns for different subgroups. Typically, for persons with a background of migration, the dimensions of material deprivation and housing contribute relatively more to multidimensional poverty, whereas health appears to contribute relatively less. However, the profiles seem to converge over time, as the dimension-specific differences decrease in general (previous year’s results not shown). Figure6 (b), however, shows that virtually any absolute contribution is larger for individuals with migration background. Similarly, fig- ure6 (a) also suggests that the relative contributions of deprivations in social participation and health increase with age, so that the roles of housing and material deprivation decrease. In absolute terms, however, figure6 (b) shows each indicator’s contribution to multidimensional poverty to increase with age.22 Thus, virtually all indicators contribute absolutely more for the elderly and for people with migration background. However, only for age do relative contributions change: in old age health and social participation become

20Naturally, marked differences in population are driving this result. State-specific adjusted headcount ratios reveal differences among states, but fail to provide clear-cut conclusions (results not shown). 21The corresponding contributions to the simple headcount ratio are presented in figure A.2. 22Figure7 shows dimensional breakdowns for the type of household and the father’s education. Note that singles and single parents exhibit remarkably similar deprivation profiles.

18 increasingly important. The fact that multidimensional poverty, M0, can be reduced to contributions of subgroups and dimensions, allows a consistent and deep analysis of multidimensional poverty, and thus a better understanding thereof.

Dynamics. The natural starting point for studying poverty dynamics is first to document changes over time. Figure8 (a) contains absolute changes in M0 for several k. Clearly, multidimensional poverty increases during the first part of the decade, and decreases during the second part. Both findings are independent of the chosen poverty cutoff k. However, (absolute) decreases are smaller for larger k. At k = 33 multidimensional poverty remains approximately unchanged after the 10 years under investigation. Moreover, figure8 (b) not only plots the relative changes of M0, but also decomposes the changes into the contributions of H and A (according to eq (3). This decomposition reveals that only for low values of k, the poverty intensity A contributes to the changes in M0. The reason is that for higher k-cutoffs individuals being relieved of one deprivation are more likely to leave poverty completely. In order to obtain a deeper insight into changes of multidimensional poverty one can compare changes in censored and uncensored headcounts, which are both depcited in figure9. 23 More specifically, the three employment indicators and the material deprivation indicator exhibit relatively high changes in both censored and uncensored headcounts during the first half of the decade. Apparently, these four indicators drive the overall increase in multidimensional poverty observed from 01/02 to 06/07. Similarly, indicators for education and unemployment play a crucial role for reducing M0 during the second half of the decade. Other patterns, however, are more difficult to rationalize and require a more carfeful analysis. For instance, in Figure9, the changes in the simple uncensored headcount of education, suggests an improvement from 01/02 to 06/07. In contrast, the censored headcount of education (at k = 33) for the same period hardly changes at all. While several different underlying trends may produce this pattern, it simply states that the same share of the population is still multidimensionally poor and deprived in education—despite the decrease in the uncensored headcount. Thus, education among the multidimensional poor calls for more attention of both policy makers and reasearchers alike. Moreover, censored headcounts also suggest that precarious employment and underemployment increase among the poor during the second half of the decade, despite the overall decrease

23 Note that weighted changes of censored headcount ratios sum up to the overall change in M0. Presenting this breakdown would however impede the direct comparison with uncensored headcount ratios.

19 in M0 for this period. Thus, despite the welcome decrease in poverty, these results caution against exaggerated optimism. A complementary analysis is to study changes by subpopulations. In principal (absolute and relative) changes in M0 can be decomposed into contributions of subpopulations. How- ever, it is importnant to correctly account for changing population shares, which would affect M0 as well. A simpler and yet instructive exercise compares absolute changes in

M0 by selected subgroups. Figure 10 shows absolute changes by four different subgroups. East-Germany, for instance, experiences both a larger (absolute) increase during the first half and a smaller decrease during the second half. Consequently, the gap in multidimensional poverty between both regions increases during the period investigated. Asymmet- ric changes can also be observed for other socio-demographic groups. Most age groups, for example, first experience increases in M0, but only youngest (<25) and oldest (65+) people are finally better off. For the intermediate age groups (25–45 and 45–65) the improvements during the second half of the decade fail to offset the worsenings of the first half. However, with respect to a background in migration things appear to be different, since migrants experienced a stronger increase during the first half and a stronger decrease during the second half of the period investigated. Moreover, these improvement finally result in a overall shrinking poverty gap between migrants and non-migrants. Fi- nally, distinguishing different household types reveals that multidimensional poverty of non-standard (i.e. “other”) household compositions increases throughout the whole period investigated—contrary to the general decrease during the second part. In summary, the Alkire-Foster framework not only documents multidimensional poverty and its changes, but also provides features to consistently detect specific patterns (e.g., changing gaps or other asymmetric impacts). Moreover, these features also allow to identify the driving factors behind changes in poverty (e.g., changes employment or material deprivation indicators). A more comprehensive analysis may not only combine the presented exercises more systematically, i.e. decomposing changes by subgroup and dimensions (or incidence-intensity-breakdown). In addition to this, an even deeper analysis of changes requires a (balanced) panel data setup, which however is beyond the scope of the present study (see Suppa, 2015a). Only then the underlying trends, which drive overall changes can be unambiguously identified, allowing an even better understanding of the mechanisms behind poverty.

Multidimensional and Income Poverty. Income-poverty is both an alternative to multidimensional poverty measures and a potential dimension. To begin with, ﬁgure 11 con-

20 trasts the respective headcount ratios for income poverty and multidimensional poverty, each for several poverty cutoffs. Note that k = 33 and 60%-of-median-income imply roughly the same incidence (ca. 11%). Moreover, monetary poverty rates are slightly increasing overtime. In particular, monetary poverty also increases from 2006/07 to 2011/12, for which multidimensional measures indicates a decrease (see also figure8). Suppa (2015a) studies changes in both measures more carefully. An important question is whether both measures identify the same individuals as poor. Naturally, such a comparison depends on the poverty cutoffs. Figure 12 shows the population shares of individuals who are considered poor (i) by both measures (both-poor), (ii) by income poverty only (IO-poor), and (iii) by multidimensional poverty only (MDO-poor). These shares are plotted for k = 27, 33, 38 and for income poverty cutoffs of 50%, and 60% of the median net household equivalence income. By construction the sum of IO-poor and both-poor is constant within a subplot. Likewise, the population shares of MDO-poor and both-poor decrease mechanically with k. For k = 33 and an income poverty cutoff of 60% only 5% of the population is identified as poor by both measures. Moreover, the shares of the IO-poor and MDO-poor are 8% and 5%. Neither other cutoffs nor different years essentially affect this finding (see also figure A.3 in the appendix). The results, therefore, suggest a substantial disagreement of both measures on who is poor. Hence, with respect to targeting the poor the choice of measure makes a difference.

Income as a dimension? The previous section revealed that multidimensional and income measures identify partly different people as poor. However, income-poverty is both an alternative to multidimensional poverty measures and a potential dimension. In fact, previous studies frequently used income as a dimension (Alkire et al., 2014, Busch and Peichl, 2010, Rippin, 2012). Importantly, including a lack-of-income dimension intro- duces the risk of double-counting. What might be counted is not a novel deprivation, but instead the income-driven lack of, e.g., health or social participation. Moreover, this paper proposes to use resource dimensions only, if their indicators—argumentatively or evidentialy—contribute extra information on otherwise ignored functionings. Then the crucial questions are: does income-poverty provide novel information about shortfalls in some functionings and to what extent do we double-count deprivations? These questions can be approached in two ways. Exploratively, who are the IO-poor, do they suffer from other deprivations and how do they differ from non-poor?24 Table4

24Studying the MDO- or both-poor in more detail may provide insights on the people ignored by income- based measures and the role of income for multidimensional poverty more generally. However, both exercises are beyond the scope of the present paper.

21 contains information about both socio-demographic background and suffered deprivations by poverty status. Evidently, IO-poor are younger than non-poor, in particular the share of individuals aged 25 or less is larger for IO-poor. Additionally, both single and single- parents are more prevalent household types among IO-poor. Turning to the deprivation the respective groups suffer, the IO-poor indeed exhibit a slightly higher average deprivation count of 0.17 compared with the non-poor (0.1), even though a much lower one than MDO- or both-poor (.4 and .45). However, most deprivation indicators are similar in size for IO-poor and non-poor. The outstanding exception are both material deprivation indicators, which are substantially higher for the IO-poor and may well explain their higher deprivation count. This finding points to a sizeable, though not surprising, overlap of material deprivation indicators with income-poverty. This overlap can also be observed pre-identification. Inspection of table5 reveals that 16% of the income-poor are materially deprived, 21% are deprived in wealth, and 34% are both. Thus 71% of the income-poor are considered deprived in at least one material deprivation indicator.25 Put differently, to the extent in which a low income translates into material deprivation or is accompanied by a lack of wealth, income-poverty is already accounted for. Thus, adding income as a dimension is likely to introduce substantial double-counting. Additionally, one may also question whether income adequately proxies even the material well-being of the IO-poor. In fact, Slesnick(2001, p.196–97), notes that in particular for young and elderly income does not accurately reflect well-being. The major reason is that income underestimates actual consumption, as the role of wealth is ignored. Wealth may not only be directly consumed, but can also provide a service flow from its stock (e.g., self-occupied property or durables). As shown by Table4, the young (aged 30 and below) are not only overrepresented among the IO-poor, moreover, together with the elderly (aged 60 and above) they account for ca. 60% of the IO-poor. Table4 also contains frequently collected wealth information. It turns out, that 25% of the IO-poor own their accommodation and 64% own a car, indicating a substantially share of this subgroup to have indeed access to wealth. Figure 13 provides more detailed wealth information by poverty status and age groups.26 The left figure shows the median of net household wealth. Notably, older IO-poor seem to have access to significant amounts of wealth. Specifically, individuals aged 45–65 report a median net household wealth of 34k EUR, whereas persons aged 65 or more report 47k EUR. To better assess the prevalence of wealth access, the right figure shows the share of individuals whose net household wealth is larger than 3500

25This pattern is also suggested by the correlation coefﬁcients in Table A.2. 26This information is collected by a comprehensive SOEP wealth module, which is however, only available in 2002 and 2007, see Frick et al. (2007).

22 EUR (the 25%-percentile of the overall wealth distribution). The results suggest that over 60% of the older IO-poor individuals own some wealth, but also 30–40% of the younger age groups. Finally, it should be noted that 23% of the IO-poor aged 25 or less currently do an apprenticeship, and another 24% is currently in some educational training. Thus, the evidence suggests that income indeed does not accurately reflect material well-being, in particular the young and the elderly. Theoretically, this is supported by the permanent income hypothesis, which implies consumption smoothing behaviour. Conceptually, one could draw on social participation to justify a lack-of-income dimension. Social participation is shaped by customs, organization and endowment of a society, which is why it is also often used to justify a relative income-poverty cutoff. Thus, if social participation was not already included as a dimension on its own, there might be a case for adding a lack-of-income dimension Table A.2, suggests income-poverty and social participation indicators to be correlated (0.16 and 0.28). Similar arguments could be made if material deprivation indicators were unavailable. Alternatively, one could argue to replace material deprivation indicators with an income- dimension. However, material deprivation indicators can be linked more closely to specific functionings, as practical reason and economic security in the present study. More generally, consumption information is often argued to be preferable to income information as it is conceptually closer to well-being.27 For instance, the role of wealth in consumption smoothing or owner-occupied property is well-established. Finally, externally fixing a uni- form income-poverty threshold, attracted also substantial critique in general (e.g., Sen, 1992, ch.7). In the present context the question is how to choose a reasonable cutoff, given other dimensions are already accounted for directly. Summing up, I reject a lack-of-income dimension based on both conceptual and empirical grounds. First, social participation, a key dimension income is important for, is already directly implemented. Second, material deprivation is also included and can be related to deprivations in two further functionings more directly (economic security and practical reason). Third, there is evidence for redundancy, as the higher deprivation of IO-poor is basically driven by material deprivation indicators. Finally, for a significant share of the IO-poor (the young and the old) income seems not to accurately reflect even their material well-being.

27In standard economic theory goods rather than income provide utility, moreover, consumption is argued to better measure permanent income.

23 5 Concluding Remarks

Outlook. Instead of another summary, I conclude with some final remarks. Better official poverty measures are feasible. By now, conceptional frameworks have been carefully de- vised, and sound and flexible methods have been developed. Many advanced economies have convenient high-quality data at hand. Importantly, both key concepts and major empirical findings are still easy to communicate and therefore may foster public debate. Moreover, a convincing poverty measure promotes the recognition of poverty as an press- ing issue more generally—in particular for an advanced economy like Germany. Thereby, it also helps to organize majorities in parliaments necessary to approve appropriate measures. Additionally, an effective and efficient fight on poverty is feasible, once the truly deprived can be better targeted, and once the coupling of deprivations and their mechanisms are better understood. Improving the lives of the most seriously deprived is within reach.

Limitations. First, complex functionings like agency, self-respect, economic security and practical reason are currently only captured indirectly. However, research on providing direct implementations already commenced (Alkire, 2007). A further aspect generally ignored is the role of time, which may, e.g., illuminate the contrasting living conditions of singles with and without children. Unfortunately, its conceptual and empirical integration is complex and still requires more thought.28 Moreover, the previous analysis of multidimensional poverty and socio-demographic variables is basically descriptive. Confounding factors may drive some findings, while certain variables are obviously highly endogenous (e.g., type of household). Consequently, these findings cannot be interpreted causally, though future research may well address this issue. Finally, given the current data, more detailed analyses of shocks and reforms are not feasible. Assuming a consensus on the relevant indicators, collecting all items on a yearly basis is, however, straight forward. Finally, certain groups of the society are ignored completely. Homeless people, for instance, are not covered by the underlying data basis. Children, on the other hand, are deliberately excluded, since a more tailored specification to accurately capture their being and doing seems called for.

Future Research. The next steps towards a multidimensional poverty index for Germany should explore the options for direct implementations of missing dimensions such as such

28Contributions approaching this issue include Merz and Rathjen(2014a,b).

24 as agency, self-respect, security, practical reason and appearance in public without shame. Additionally, a clear conceptual account of both employment and time is a precondition for a better empirical integration of employment- and time-related deprivation into poverty measures. Regarding the data basis, having yearly data of all indicators would allow a more detailed analysis. Questions on Internet-based social participation may complement the present implementation. Finally, applying methods that take account of confounding factors would deepen the analysis and help to uncover the mechanism behind multidimensional poverty.

References

Alkire, S., 2007. The Missing Dimensions of Poverty Data: Introduction to the Special Issue, Oxford Development Studies, 35 (4), 347–359.

Alkire, S., Apablaza, M., and Jung, E., 2014. Multidimensional poverty measurement for EU-SILC countries, OPHI Research in Progress Series 36b, OPHI, Oxford.

Alkire, S., Ballon, P.,Foster, J., Roche, J.M., Santos, M.E., and Seth, S., 2015. Multidimen- sional Poverty Measurement and Analysis: A Counting Approach, Oxford Univ. Press.

Alkire, S. and Foster, J., 2011a. Counting and Multidimensional Poverty Measurement, Journal of Public Economics, 95 (7-8), 476–487.

Alkire, S. and Foster, J., 2011b. Understandings and Misunderstandings of Multidimen- sional Poverty Measurement, Journal of Economic Inequality, 9 (2), 289–314.

Alkire, S. and Foster, J.E., 2013. Evaluating dimensional and distributional contributions to multidimensional poverty, mimeo, OPHI, University of Oxford.

Alkire, S. and Santos, M.E., 2011. Acute Multidimensional Poverty: A New Index for De- veloping Countries, Proceedings of the German Development Economics Conference, Berlin 2011 3, Verein für Socialpolitik, Research Committee Development Economics, Berlin.

Allardt, E., 1993. Having, Loving, Being: An Alternative to the Swedish Model of Welfare Research, in: A.K. Sen and M.C. Nussbaum, eds., The Quality of Life, Clarendon Press, 88–94.

25 Arndt, C. and Volkert, J., 2007. A Capability Approach for Ofﬁcial German Poverty and Wealth Reports: Conceptual Background and First Empirical Results, IAW - Discussion Papers 27, Institut für Angewandte Wirtschaftsforschung, Tübingen.

Atkinson, A.B., Marlier, E., and Nolan, B., 2004. Indicators and Targets for Social Inclusion in the European Union, Journal of Common Market Studies, 42 (1), 47–75.

Atkinson, T., Cantillon, B., Marlier, E., and Nolan, B., 2002. Social Indicators: The EU and Social Inclusion, Oxford: Oxford University Press.

Bourguignon, F.and Chakravarty, S., 2003. The Measurement of Multidimensional Poverty, Journal of Economic Inequality, 1 (1), 25–49.

Bundesregierung, 2008. Lebenslagen in Deutschland., 3. Armuts- und Reichtumsbericht, BMAS, Berlin.

Bundesregierung, 2013. Lebenslagen in Deutschland, 4. Armuts- und Reichtumsbericht, BMAS, Bonn.

Busch, C. and Peichl, A., 2010. The Development of Multidimensional Poverty in Germany 1985-2007, IZA Discussion Papers 4922, Institute for the Study of Labor (IZA).

Csíkszentmihályi, M., 1990. Flow: The Psychology of Optimal Experience, New York: Harper and Row.

Duclos, J.Y., Sahn, D.E., and Younger, S.D., 2006. Robust Multidimensional Poverty Com- parisons, Economic Journal, 116 (514), 943–968.

Erikson, R., 1993. Descriptions in Inequality: The Swedish Approach to Welfare Research, in: A.K. Sen and M.C. Nussbaum, eds., The Quality of Life, Clarendon Press, 67–83.

Evans, M., Kelley, J., Sikora, J., and Treiman, D.J., 2010. Family scholarly culture and educational success: Books and schooling in 27 nations, Research in Social Stratiﬁcation and Mobility, 28, 171–197.

Ferreira, F.H.G., 2011. Poverty Is Multidimensional. But What Are We Going to Do About It?, Journal of Economic Inequality, 9 (3), 493–495.

Foster, J., Greer, J., and Thorbecke, E., 1984. A Class of Decomposable Poverty Measures, Econometrica, 52 (3), 761–66.

26 Frick, J.R., Grabka, M.M., and Marcus, J., 2007. Editing and multiple imputation of item- non-response in the 2002 wealth module of the german socio-economic panel (soep), SOEPpapers on Multidisciplinary Panel Data Research 18, DIW Berlin, Berlin.

Friedman, M., 1956 [2015]. A Theory of the Consumption Function, Martino Fine Books.

Goebel, J., Grabka, M.M., Krause, P., Kroh, M., Pischner, R., Sieber, I., and Spieß, M., 2008. Mikrodaten, Gewichtung und Datenstruktur der Längsschnittstudie Sozio- oekonomisches Panel (SOEP), Vierteljahrshefte zur Wirtschaftsforschung, 77 (3), 77–109.

Hahn, M. and Haisken-DeNew, J.P.,2013. PanelWhiz and the Australian Longitudinal Data Infrastructure in Economics, The Australian Economic Review, 46 (3), 379–386.

Haisken-DeNew, J.P.and Hahn, M., 2010. PanelWhiz: Efﬁcient Data Extraction of Complex Panel Data Sets: An Example Using the German SOEP, Schmollers Jahrbuch, 130 (4), 643–654.

Hetschko, C., Knabe, A., and Schöb, R., 2013. Changing Identity: Retiring from Unemploy- ment, Economic Journal, 124 (575), 149–166.

Kunze, L. and Suppa, N., 2014. Bowling Alone or Bowling at All? The Effect of Unemploy- ment on Social Participation, Ruhr Economic Papers 510, RWI, Essen.

Lazarsfeld, P.F.,1958. Evidence and Inference in Social Research, Daedalus, 87 (4), 99–130.

Leßmann, O., 2009. Conditions of Life, Functionings, and Capability - Similarities and Differences, Journal of Human Development and Capabilities, 10 (2), 279–298.

Leßmann, O. and Bonvin, J.M., 2011. Job-Satisfaction in the Broader Framework of the Capability Approach, management revue, 22 (1), 84–99.

Mani, A., Mullainathan, S., Shaﬁr, E., and Zhao, J., 2013. Poverty Impedes Cognitive Func- tion, Science, 341, 976–980.

Marlier, E. and Atkinson, A.B., 2010. Indicators of Poverty and Social Exclusion in a Global Context, Journal of Policy Analysis and Management, 29 (2), 285–304.

Merz, J. and Rathjen, T., 2014a. Multidimensional time and income poverty: well-being gap and minimum 2DGAP poverty intensity – German evidence, Journal of Economic Inequality, 12 (4), 555–580.

27 Merz, J. and Rathjen, T., 2014b. Time and Income Poverty: An Interdependent Multidi- mensional Poverty Approach with German Time Use Diary Data, Review of Income and Wealth, 60 (3), 450–479.

Mullainathan, S. and Shaﬁr, E., 2013. Scarcity: Why Having Too Little Means So Much, London: Allen Lane.

Neurath, O., 1917 [2006]. Das Begriffsgebäude der Wirtschaftslehre und seine Gurndlagen, Zeitschrift für die gesamte Staatswissenschaft, 73, 484–520, reprinted in Uebel, T. and Cohen, R. (eds.): Otto Neurath: Economic Writings: Selections 1904-1945, 2006, pp. 312-341.

Neurath, O., 1937 [2006]. Inventory of the Standard of Living, Zeitschrift für Sozial- forschung, 6, 140–151, reprinted in Uebel, T. and Cohen, R. (eds.): Otto Neurath: Eco- nomic Writings: Selections 1904-1945, 2006, pp. 513-526.

Nussbaum, M.C., 2001. Women and Human Development: The Capabilities Approach, The John Robert Seeley Lectures, vol. 1998, Cambridge: Cambridge University Press, 13th ed.

OECD, 2011. How’s Life? Measuring Well-being: Measuring Well-being, OECD Better Life Initiative, OECD Publishing.

OECD, 2014. Education at a Glance, OECD Indicators, Paris: OECD Publishing.

Pattanaik, P.K. and Xu, Y., 1990. On Ranking Opportunity Sets in Terms of Freedom of Choice, Discussion Papers (REL - Recherches Economiques de Louvain) 1990036, Uni- versité catholique de Louvain, Institut de Recherches Economiques et Sociales (IRES).

Rippin, N., 2012. Operationalising the Capability Approach: A German Correlation Sen- sitive Poverty Index, Discussion Papers 132, Courant Research Centre, Georg-August- Universität Göttingen.

Robeyns, I., 2005. The Capability Approach: A Theoretical Survey, Journal of Human De- velopment, 6 (1), 93–117.

Roche, J., 2013. Monitoring Progress in Child Poverty Reduction: Methodological Insights and Illustration to the Case Study of Bangladesh, Social Indicators Research, 112 (2), 363–390.

Schmitz, H., 2011. Why are the unemployed in worse health? The causal effect of unemployment on health, Labour Economics, 18 (1), 71–78.

28 Sen, A.K., 1985. Commodities and Capabilities, New Delhi: North-Holland Publ., 12th ed.

Sen, A.K., 1992. Inequality Reexamined, Russell Sage Foundation book, New York: Russell Sage Foundation, 3rd ed.

Sen, A.K., 1999a. Democracy as a Universal Value, Journal of Democracy, 10 (3), 3–17.

Sen, A.K., 1999b. Development as Freedom, Oxford: Oxford University Press.

Sen, A.K., 2004. Capabilities, Lists and Public Reason: Continuing the Conversation, Femi- nist Economics, 10 (3), 77–80.

Shah, A., , Mullainathan, S., and Shaﬁr, E., 2012. Some Consequences of Having Too Little, Science, 338 (2), 682–685.

Slesnick, D.T., 2001. Consumption and Social Welfare, Cambridge, UK: Cambridge Univ. Press.

Stiglitz, J.E., Sen, A.K., and Fitoussi, J.P., 2009. Report by the commission on the measurement of economic performance and social progress, Tech. rep., Commission on the Measurement of Economic Performance and Social Progress.

Suppa, N., 2014. The Capability Perspective: Basic Features and their Relevance for So- cial Policy, in: H.U. Otto and H. Ziegler, eds., Critical Social Policy and the Capability Approach, Verlag Barbara Budrich.

Suppa, N., 2015a. Comparing monetary and multidimensional poverty in Germany, mimeo, TU Dortmund, Dortmund.

Suppa, N., 2015b. Labor and the Capability Approach. Towards Conceptional Clarity, mimeo, TU Dortmund, Dortmund.

Townsend, P., 1979. Poverty in the United Kingdom: A Survey of Household Resources and Standards of Living, University of California Press.

Tsui, K., 2002. Multidimensional Poverty Indices, Social Choice and Welfare, 19 (1), 69–93.

UN, 2012. Millennium Development Goals Report: 2012, Millennium Development Goals Reports, New York: United Nations Publications.

UNDP,1990. Human Development Report 1990: Concept and Measurement of Human Devel- opment, Human Development Report, Oxford University Press.

29 UNDP,2011. Human Development Report 2011: Sustainability and Equity: Towards a Better Future for All, Human Development Report, Palgrave Macmillan.

Wagner, G.G., Frick, J.R., and Schupp, J., 2007. The German Socio-Economic Panel Study (SOEP): Scope, Evolution and Enhancements, Schmollers Jahrbuch, 127 (1), 139–169.

Whelan, C.T., Nolan, B., and Maître, B., 2014. Multidimensional Poverty Measurement in Europe: An Application of the Adjusted Headcount Approach, Journal of European Social Policy, 24, 183–197.

WHO, 2000. Obesity: Preventing and Managing the Global Epidemic, no. 894 in World Health Organization technical report series, Geneva: World Health Organization.

Winkelmann, L. and Winkelmann, R., 1998. Why Are the Unemployed So Unhappy? Evi- dence from Panel Data, Economica, 65 (257), 1–15.

Wissenschaftszentrum Berlin für Sozialforschung (WZB) and Institut für Arbeitsmarkt- und Berufsforschung (IAB), 2013. Soziale Mobilität, Ursachen für Auf- und Abstiege, on be- half of BMAS (ed.), Bonn.

Wolff, J. and de-Shalit, A., 2007. Disadvantage, Oxford Political Theory, Oxford: Oxford University Press.

Yitzhaki, S., 1979. Relative Deprivation and the Gini Coefﬁcient, The Quarterly Journal of Economics, 93 (2), 321–24.

30 Table 1: Potential Dimensions

dimension NB ACMN SSF OECD RPW functioning education health housing () (shelter, health, privacy) social participation political participation agency practical reason () self-respect employment () () (but also self-respect, agency) income (multipurpose) material deprivation () (numerous, depends on items) environment aspects (health, shelter) time (activities) (multipurpose) security ()() () () (secure functionings?) Notes: NB is Nussbaum(2001), ACMN is Atkinson et al. (2002), SSF is Stiglitz et al. (2009), OECD is OECD(2011)

31 Table 2: Functionings, Indicators, and Weights

Functioning Deprivation Cut-off Variable Weight

elementary schooling not completed or elementary schooling completed but a Education no vocational qualiﬁcation dep_educ 1/12 less than 10 books in household dep_nbooks 1/12

house requires major renovation or is ready for demolition dep_housecond 1/18 Housing neither of bath or shower, kitchen, warm water, toilet dep_hhfacilities 1/18 overcrowded (less than one room per person) dep_overcrowded 1/18

partially or severely disabled dep_disability 1/18 b Health reporting 2/4 health issues dep_healthidx 1/18 body mass index larger than 30 dep_obesity 1/18

c 32 reporting 2 4 goods missing for ﬁnancial reasons dep_matdep 1/18 Material / none of life insurance, pension, owning the house or apartment, other dep_wealth 1/18 Deprivation house, ﬁnancial assets, commercial enterprise, tangible assets

d Social 5/7 activities performed never ; remaining at most less than monthly dep_actindex 1/12 Participation never meeting friends dep_meetfriends 1/12

unemployed dep_unemp 1/6 Employment invol. hours worked < 30 dep_underemp 1/18 precariously employed (incl. temporary work ) dep_precemp 1/18

Notes: Note: aGraduation in Germany is usually achieved after 10 years of schooling. bThe four health issues are (i) a strong limitation when climbing stairs, (ii) a strong limitation for tiring activities, (iii) physical pain occured always or often during the last 4 weeks, and (iv) the health condition limited always or often socially. cThe four goods asked for are (i) a warm meal, (ii) whether friends are invited for dinner, (iii) whether money is put aside for emergencies, and (iv) whether worn out furniture is replaced. d Activities included are (i) going to the movies, pop music concerts, dancing, disco, etc, (ii) going to cultural events (such as concerts, theater, lectures), (iii) doing sports yourself, (iv) volunteer work, (v) attending religious events, (vi) helping out friends, relatives or neighbours (vii) involvement in a citizens’ group, political party, local government. Table 3: Deprivation headcount ratios uncensored headcount censored headcount share non-poor depr. non-poor m-poor total dep_educ 0.085 0.471 0.126 0.050 0.604 dep_Nbooks 0.032 0.292 0.060 0.031 0.481 dep_healthidx 0.116 0.409 0.147 0.043 0.704 dep_disability 0.124 0.319 0.145 0.034 0.766 dep_obesity 0.169 0.372 0.190 0.039 0.793 dep_housecond 0.017 0.082 0.024 0.009 0.634 dep_overcrowded 0.042 0.134 0.052 0.014 0.725 dep_hhfacilities 0.010 0.039 0.013 0.004 0.676 dep_matdep 0.115 0.669 0.174 0.071 0.591 dep_wealth 0.176 0.789 0.241 0.084 0.653 dep_actindex 0.172 0.627 0.220 0.067 0.697 dep_meetfriends 0.188 0.540 0.225 0.057 0.745 dep_unemp 0.010 0.320 0.043 0.034 0.208 dep_underemp 0.057 0.100 0.062 0.011 0.827 dep_precemp 0.061 0.107 0.066 0.011 0.828 Notes: Data from SOEP v29.1. Calculations for 2012, cells contains shares. Underlying poverty cutoff k = 33.

Figure 1: Aggregate Measures over Time

M0 H A .08 .55 .2 .06 .5 .15 .04 .45 .1 .4 .02 .05 0 0 .35 25 30 35 40 45 50 25 30 35 40 45 50 25 30 35 40 45 50 k k k

2001−02 2006−07 2011−12

Note: dahsed line at k=33.33

Notes: Data from SOEP v29.1.

33 Table 4: Statistics by poverty status (1) (2) (3) (4) non-poor both-poor IO-poor MDO-poor age <25 0.08 0.08 0.21 0.06 25-30 0.07 0.08 0.10 0.05 31-39 0.15 0.13 0.12 0.12 40-49 0.20 0.21 0.16 0.16 50-59 0.16 0.23 0.13 0.20 60-69 0.16 0.14 0.12 0.18 70+ 0.17 0.12 0.16 0.24 hh-type single 0.22 0.36 0.33 0.33 couple, no kids 0.37 0.19 0.19 0.31 single-parent 0.04 0.13 0.10 0.08 couple w. kids 0.35 0.29 0.33 0.26 other 0.02 0.03 0.04 0.03 deprivations dep_educ 0.10 0.49 0.17 0.52 dep_Nbooks 0.03 0.37 0.07 0.28 dep_disability 0.12 0.24 0.09 0.36 dep_obesity 0.15 0.30 0.15 0.36 dep_healthidx 0.11 0.33 0.12 0.46 dep_housecond 0.02 0.11 0.05 0.08 dep_overcrowded 0.04 0.18 0.12 0.11 dep_hhfacilities 0.01 0.06 0.04 0.06 dep_unemp 0.02 0.44 0.05 0.24 dep_underemp 0.05 0.08 0.08 0.07 dep_precemp 0.05 0.08 0.09 0.08 dep_wealth 0.16 0.84 0.40 0.70 dep_matdep 0.09 0.77 0.35 0.50 dep_meetfriends 0.19 0.47 0.20 0.55 dep_actindex 0.19 0.63 0.23 0.69 house owner 0.43 0.08 0.24 0.12 car owner 0.88 0.39 0.64 0.60 share indebted 0.06 0.14 0.08 0.16 counting vector 0.10 0.45 0.17 0.40 N 40537 1881 3075 2617 Notes: Data from SOEP v29.1. Waves 2001/02, 2006/07, 2011/12. Cells contain shares. Underlying k-cutoff is 33%, income-poverty cutoff is 60%.

34 Table 5: Cross-tabulation of income-poor and material deprivation indicators non-income-poor income-poor

dep_mat=0 dep_mat=1 Total dep_mat=0 dep_mat=1 Total dep_wealth=0 75.83 6.42 82.25 28.80 16.22 45.03 dep_wealth=1 13.54 4.21 17.75 20.87 34.10 54.97 Total 89.37 10.63 100.00 49.67 50.33 100.00 Notes: Data SOEP v29.1, cells contain percentages, income poverty cutoff is 60% of median dis- posable household equivalence income. Deprivation indicators are deﬁned as in table2.

35 Figure 2: Aggregate Measures by Subgroups I (a) migration background .6 .15 .4 .1 0 H M .2 .05 0 0 10 20 30 40 50 10 20 30 40 50 k k

no migr. migr.

(b) region .15 .5 .4 .1 .3 0 H M .2 .05 .1 0 0 10 20 30 40 50 10 20 30 40 50 k k

west east

<25 25−45 45−65 65<

Notes: Data from SOEP v29.1. Calculations for 2011/2012.

36 Figure 3: Aggregate Measures by Subgroups II (a) type of household .2 .8 .6 .15 0 H .1 .4 M .2 .05 0 0 10 20 30 40 50 10 20 30 40 50 k k

single couple, no kids single−parent couple w. kids other

(b) education of father .8 .2 .6 .15 0 H .4 M .1 .2 .05 0 0 10 20 30 40 50 10 20 30 40 50 k k

not completed or d.k. Hauptschule Realschule Abitur + other

men women

Notes: Data from SOEP v29.1. Calculations for 2011/2012.

37 Figure 4: Contributions to M0 by Subgroup migration background region

27.37% 32.11%

no migr. west migr. east

67.89% 72.63%

Note: k=33, period of analysis: 2011−12 Note: k=33, period of analysis: 2011−12 education of father state (Bundesland)

BW 3.24% 4.91% 4.36% 2.65% 8.78% BAV 4.82% BER 7.51% 27.95% BRA 5.33% 14.45% BRE 1.81% HAM not completed or d.k. 4.33% HES Hauptschule MV Realschule 3.42% LS Abitur + NRW other 7.11% RP

0.92% SAA 1.18% 25.81% SAX 6.15% SA 56.94% 2.46% 5.87% SH THU

Note: k=33, period of analysis: 2011−12 Note: k=33, period of analysis: 2011−12 Notes: Data from SOEP v29.1. Year of analysis 2011/12. Graphs show contribution to overall M0 for selected subgroups, poverty cutoff k = 33. For comparison only: share of population with migration background 19%.

38 Figure 5: Dimensional Breakdown

by period (abs. contr.) by period (rel. contr.)

0.1 0.8 1.2 0.1 0.7 1.0 1.5 100 1.4 5 dep_educ_cont 11.8 0.7 13.8 11.9 dep_Nbooks_cont 0.1 0.0 0.1 10.7 dep_healthidx_cont 0.5 10.7 0.5 0.6 80 9.8 4 dep_disability_cont

0.5 0.5 14.6 0.7 13.2 12.6 dep_obesity_cont

dep_housecond_cont

0.6 0.6 60 3

0.8 15.1 14.9 15.9 dep_overcrowded_cont

0.7 0.7 dep_hhfacilities_cont

10.2 0.7 13.0 13.2 dep_matdep_cont 40 2 0.5 1.2 0.6 2.0 0.0 0.6 0.5 dep_wealth_cont 0.1 1.2 1.7 1.8 0.1 0.1 4.1 1.2 1.0 0.1 0.0 0.1 0.2 0.1 4.4 4.9 0.2 0.0 4.2 dep_actindex_cont 0.2 0.2 0.2 3.9 4.3 5.5

1 0.2

20 dep_meetfriends_cont 0.2 0.3 5.2 5.5 0.2 7.0 0.3 0.3 6.3 5.5 dep_unemp_cont 0.3

11.1 dep_underemp_cont 0.5 0.5 0.4 9.6 9.3

0 0 dep_precemp_cont

2001−02 2006−07 2011−12 2001−02 2006−07 2011−12

Note: k=33, absolute contributions to M0 are multiplied by 100.

Note: Data from SOEP v29.1. Poverty cutoff k = 33. For better readability all (weighted) contributions are multiplied by 100. Thus, relative contributions are percentage points, whereas absolute contribution sum to M0 100. ×

39 Figure 6: Dimensional Breakdown by Subpopulations I

(a) relative contribution M0

by age by migration background

1.1 0.7 3.6 3.2 0.1 2.2 3.5 100 100 dep_educ_cont

15.3 12.5 11.5 20.3 14.4 dep_Nbooks_cont 18.0 dep_healthidx_cont 11.6 80 13.1 80 13.4 dep_disability_cont 8.0 9.4 19.1

14.1 dep_obesity_cont 8.8 14.0 12.3 12.9 dep_housecond_cont

10.1 60 60

14.0 0.7 9.9 11.7 dep_overcrowded_cont 11.4 9.8 5.5 1.2 1.9 1.8 3.0 0.80.2 dep_hhfacilities_cont 2.4 1.9 8.2 6.7 9.2

9.8 40 0.6 dep_matdep_cont 40 9.5 9.4 1.2 1.2 0.6 0.2 7.0 4.0 0.3 0.6 6.1 5.1 1.1 1.0 dep_precemp_cont 5.0 1.3 1.3 5.2 1.2 6.2 5.4 3.6 10.4 3.4 dep_wealth_cont 1.6 4.8 1.0 5.5 2.2 7.0 5.0 20 dep_actindex_cont 20 6.4 2.9 8.3 6.7 5.2 6.7 5.9 dep_meetfriends_cont 5.3 15.9 13.9 10.0 12.1 11.5 dep_unemp_cont 7.9 0

0 dep_underemp_cont <25 65< migr. 25−45 45−65 no migr. period: 2011/12 period: 2011/12

Note: k=30, absolute contributions to M0 are multiplied by 100.

(b) absolute contribution M0

by age by migration background

0.1 0.1 10

6 0.1 dep_educ_cont

1.1 1.2 dep_Nbooks_cont

dep_healthidx_cont

0.0 8 0.7 0.6 0.1 0.7 dep_disability_cont 0.1 0.1 0.1 1.2 dep_obesity_cont 0.8 0.8 4 0.9 0.8

6 dep_housecond_cont 0.3 1.6 dep_overcrowded_cont 0.4 0.9 0.5 0.7 dep_hhfacilities_cont 0.5 0.00.1

0.4 4 1.3 0.8 0.6 dep_matdep_cont 0.8 0.00.1 0.7 0.5 0.0 0.3 0.4 dep_wealth_cont 2 0.10.0 0.1 0.1 0.6 0.7 0.3 0.4 0.3 dep_actindex_cont 0.3 0.0 0.7 0.6

0.2 0.3 0.5 2 0.4 0.1 dep_meetfriends_cont 0.1 0.0 0.5 0.10.0 0.1 0.3 0.7 0.1 0.10.0 0.3 0.1 0.5 0.2 dep_unemp_cont 0.1 0.3 0.2 0.1 0.2 0.2 0.3 1.1 0.7 0.7 dep_underemp_cont 0.5 0.4 0.3 0

0 dep_precemp_cont <25 65< migr. 25−45 45−65 no migr. period: 2011/12 period: 2011/12

Note: k=33, absolute contributions to M0 are multiplied by 100.

40 Figure 7: Dimensional Breakdown by Subpopulations II

(a) relative contribution M0

by hh−type by education of father

0.5 1.6 2.3 2.8 2.4 2.8 2.1 2.7 3.7 100 4.8 100 dep_educ_cont 11.1 8.7 12.4 11.2 11.0 15.3 14.0 22.2 18.7 dep_Nbooks_cont 22.1

11.3 15.5 11.0 13.4 14.4 80 dep_healthidx_cont 80 12.5 12.7 10.5 dep_disability_cont 8.4 14.0 12.3 14.6 14.4 12.2 16.2 12.1 13.6 dep_obesity_cont 9.8 12.7 60 12.0 60 12.0 dep_housecond_cont 10.8 9.5 9.7 9.6 12.5 9.9 11.5 1.7 1.4 10.8 0.3 2.0 dep_overcrowded_cont 10.9 1.1 2.1 9.4 8.0 9.0 1.6 10.2 3.5 2.7 1.7 9.0 6.4 40 0.6 1.40.6 dep_hhfacilities_cont 1.3 0.3 0.1 2.8 1.1 9.2 40 1.10.2 1.00.4 10.6 0.9 1.3 9.1 6.4 6.1 9.6 5.2 5.4 5.4 7.1 0.4 dep_matdep_cont 0.1 0.2 1.0 1.0 3.5 0.5 3.1 4.4 3.8 5.6 2.5 6.2 1.4 6.0 1.4 3.4 6.2 5.3 5.0 dep_precemp_cont 4.8 1.3 4.1 7.3 5.5 5.6 2.3 7.2 2.8 5.7 20 8.0 6.5 6.7 3.5 3.5 4.9 3.5 dep_wealth_cont 20 6.4 3.2 2.8 5.2 3.0 9.2 5.5 6.2 4.1 6.0 5.3 13.1 3.0 2.4 10.2 11.4 dep_actindex_cont 6.6 6.3

9.6 11.8 11.3 10.7 11.9 0 dep_meetfriends_cont

0 other Abitur + dep_unemp_cont other Realschule single Hauptschule dep_underemp_cont single−parent couple, no kids couple w. kids not completed or d.k. period: 2011/12 period: 2011/12

Note: k=30, absolute contributions to M0 are multiplied by 100.

(b) absolute contribution M0

by hh−type by education of father

0.0 0.3

10 dep_educ_cont 1.2 1.3 8 dep_Nbooks_cont 0.2 1.1 1.2 dep_healthidx_cont 8 1.7 dep_disability_cont 1.4

6 0.1 1.1 dep_obesity_cont 0.7 0.6 6 1.2 dep_housecond_cont 0.7 0.8 0.8 0.0 0.2 dep_overcrowded_cont 0.9 1.0 4 0.9 0.9 0.2

4 dep_hhfacilities_cont 0.0 0.1 0.1 0.7 0.1 0.3 0.5 0.6 0.5 0.1 0.6 0.5 0.5 dep_matdep_cont 0.1 0.8 0.1 0.1 0.1 0.4 0.4 0.6 0.5 0.3 0.0 0.4 0.6 0.1 0.2 0.4 0.5 dep_precemp_cont 0.10.0 0.4 0.1 0.4 0.4 0.4 0.3 0.4 2 0.1 0.0 0.1 2 0.7 0.4 0.4 0.4 0.6 0.3 0.4 0.4 0.2 0.00.1 0.1 dep_wealth_cont 0.2 0.2 0.00.2 0.3 0.3 0.3 0.2 0.0 0.3 0.4 0.1 0.1 0.6 0.2 0.2 0.2 0.3 0.00.2 0.1 0.1 0.0 0.5 0.3 1.3 0.2 0.1 0.2 0.2 0.0 0.3 0.0 0.00.1 0.1 dep_actindex_cont 0.6 0.2 0.1 0.00.1 0.1 0.0 0.4 0.1 0.00.1 0.5 0.2 0.1 0.10.0 0.2 0.1 0 0.9 0.1 1.0 0.6 0.2 0.1 dep_meetfriends_cont 0.3 0.3

0 other Abitur + dep_unemp_cont other Realschule single Hauptschule dep_underemp_cont single−parent couple, no kids couple w. kids not completed or d.k. period: 2011/12 period: 2011/12

Note: k=30, absolute contributions to M0 are multiplied by 100.

41 Figure 8: Documenting changes in M0 (a) absolute changes in M0

01/02 → 06/07 06/07 → 11/12 .01 .005 0 M0 ∆ −.005 −.01 11 13 16 19 22 25 27 30 33 36 38 41 44 47 50 11 13 16 19 22 25 27 30 33 36 38 41 44 47 50 k−cutoff Graphs by change

(b) decomposed relative changes of M0.

01/02 → 06/07 06/07 → 11/12 .6 .4 .2 0 M δ 0 −.2 −.4 11 13 16 19 22 25 27 30 33 36 38 41 44 47 50 11 13 16 19 22 25 27 30 33 36 38 41 44 47 50 k−cutoff

δH δA δHδA

Graphs by change

Notes: Data from SOEP v29.1.

42 Figure 9: Percentage changes in censored and uncensored deprivation headcounts

33 uncensored headcount

0.01 −0.14 0.02 −0.02 0.11 −0.06 0.09 −0.02 0.26 0.16 DCH_dep_educ 0.26 0.08 06/07 06/07 0.01 −0.18 DCH_dep_Nbooks −0.40 −0.48 → → 0.62 DCH_dep_healthidx 0.48 0.16 −0.02 DCH_dep_disability 0.06 −0.09 0.08 −0.02 DCH_dep_obesity 0.37 0.12 DCH_dep_housecond 0.75 0.47 0.80 0.58 DCH_dep_overcrowded

43 DCH_dep_hhfacilities DCH_dep_matdep −0.16 −0.14 −0.18 DCH_dep_wealth −0.21 −0.10 0.04 DCH_dep_actindex −0.05 0.04 −0.03 0.12 DCH_dep_meetfriends −0.30 −0.20 11/12 01/02 DCH_dep_unemp 11/12 01/02 −0.14 −0.07 −0.29 −0.16 → DCH_dep_underemp → −0.12 −0.14 −0.09 −0.00 DCH_dep_precemp −0.18 −0.10 −0.05 0.02 06/07 06/07 −0.23 −0.31 0.15 0.06 0.05 0.01

−.5 0 .5 1 −.5 0 .5 1

Notes: Data from SOEP v29.1. Left ﬁgure shows relative changes in censored headcounts at k = 33, right ﬁgure relative changes of simple uncensored headcounts. Table A.1: Questions

Activities: Which of the following activities do you take part in during your free time? Please check off how often you do each activity: at least once a week, at least once a month, less often, never.

Going to the movies, pop music concerts, dancing, disco, sports events Going to cultural events (such as concerts, theater, lectures, etc.) Doing sports yourself Volunteer work in clubs or social services Attending church, religious events Meeting with friends, relatives or neighbors Helping out friends, relatives or neighbors

44 Figure 10: Absolute changes in M0 by subgroups

regions age groups 33 33 0.0106 .02 .01 0.0136 0.0064 0.0114 0.0075 .01 .005 0 0 M ∆ M 0 ∆ 0 −0.0028 −0.0025 −0.0062 −.01 −0.0046 −0.0127 −.005

−0.0075 −0.0184 −.02 01/02 → 06/07 06/07 → 11/12 −.01 01/02 → 06/07 06/07 → 11/12 <25 25−45 west east 45−65 65<

background of migration hh−type 33 33 0.0201

.02 0.0178 .02 0.0138 0.0135

.01 0.0082 0.0082

0.0056 .01 0

M 0.0039 0 ∆ M 0 ∆ 0 −0.0004 −0.0047 −.01 −0.0081

−.01 −0.0100−0.0107 01/02 → 06/07 06/07 → 11/12 −.02 −0.0225 01/02 → 06/07 06/07 → 11/12 single couple, no kids single−parent couple w. kids no migr. migr. other

Notes: Data from SOEP v29.1. Poverty cutoff k = 33.

Figure 11: Income and Multidimensional Poverty Rates

2001−02 2006−07 2011−12 .3 .2 .1 0 20 30 40 20 30 40 20 30 40 k

50% of ymed 60% of ymed multidimensional poverty

Notes: Data from SOEP v29.1. Underlying income concept is real net household equivalence income.

45 Figure 12: Income and Multidimensional Poverty

50% of y med 60% of y med

.25 0.09

0.12

.2 0.05

0.03 0.08

.15 0.07 0.08 0.10

0.04 .1

0.03 0.04 0.05 0.07 0.05 0.04 0.04 .05 0.03 0.02 0 27 33 38 27 33 38 k−cutoff

both−poor IO−poor MDO−poor

period of analysis: 2011−12; y is real net household equivalence income.

Notes: Data from SOEP v29.1. Underlying income concept is real net household equivalence income. Year of analysis is 2011–12.

Figure 13: Wealth statistics by poverty status

median net wealth share wealthy 150000 .8 .6 100000 .4 50,000 .2 0 0 non−poor both−poor IO−poor MDO−poor non−poor both−poor IO−poor MDO−poor

<25 25−45 45−65 65<

Notes: Data from SOEP v29.1. Wave 2002, 2007. Right ﬁgure considers an individual wealthy if her wealth is more than 3.500EUR, the 25%-percentile of the entire wealth distribution.

46 Table A.2: Correlation of deprivation indicators dep_unemp dep_underemp dep_precemp dep_educ dep_Nbooks dep_disability dep_obesity dep_healthidx dep_housecond dep_overcrowded dep_hhfacilities dep_wealth dep_matdep dep_meetfriends dep_actindex pov60 dep_educ 1.000 dep_Nbooks 0.437 1.000 dep_disability 0.154 0.094 1.000

47 dep_obesity 0.142 0.092 0.189 1.000 dep_healthidx 0.276 0.203 0.668 0.280 1.000 dep_housecond 0.163 0.207 0.064 0.081 0.127 1.000 dep_overcrowded 0.232 0.139 -0.221 -0.022 -0.172 0.185 1.000 dep_hhfacilities 0.184 0.324 0.068 0.073 0.165 0.330 -0.045 1.000 dep_unemp 0.163 0.252 0.016 0.091 0.075 0.261 0.205 0.163 1.000 dep_underemp -0.017 -0.084 -0.202 -0.035 -0.168 0.068 0.087 -0.043 -1.000 1.000 dep_precemp -0.009 -0.018 -0.142 -0.026 -0.119 0.094 0.088 -0.027 -1.000 0.709 1.000 dep_wealth 0.349 0.399 0.042 0.051 0.138 0.296 0.314 0.187 0.399 0.067 0.119 1.000 dep_matdep 0.231 0.342 0.055 0.118 0.152 0.342 0.306 0.141 0.462 0.153 0.148 0.498 1.000 dep_meetfriends 0.133 0.179 0.201 0.101 0.264 0.090 -0.022 0.148 0.113 -0.047 -0.058 0.091 0.198 1.000 dep_actindex 0.375 0.428 0.243 0.148 0.348 0.198 0.137 0.158 0.198 -0.071 -0.058 0.298 0.275 0.215 1.000 pov60 0.329 0.417 0.047 0.098 0.140 0.325 0.328 0.247 0.538 0.116 0.133 0.517 0.609 0.157 0.278 1.000 Notes: Data from SOEP v29.1. Waves 2001/02, 2006/07, 2011/12. Matrix contains tetrachoric correlation coefﬁcients. pov60 indicates whether an individual is income-poor, i.e. her real net household equivalence income is less than 60% of its median. Figure A.1: Average intensity by groups .6 .6 .5 .5 A A .4 .4 .3 .3 .2 .2 10 20 30 40 50 10 20 30 40 50 k k

single couple, no kids not completed or d.k. Hauptschule single−parent couple w. kids Realschule Abitur + other other

Notes: Data from SOEP v29.1. Calculations for 2011/12.

Figure A.2: Contributions to H by Subgroups state (Bundesland) migration background

4.72% 9.11% 2.78% BAV 4.82% BER BRA 5.33% 14.68% BRE 32.38% 1.75% HAM

4.11% HES MV no migr. 3.54% LS migr. NRW 6.82% RP

SAA 67.62% 0.97% 1.21% 25.06% SAX 6.11% SA 2.64% 6.35% SH THU

Note: k=33, period of analysis: 2011−12 Note: k=33, period of analysis: 2011−12 region education of father

3.41% 4.65%

7.42% 27.19% 27.40%

not completed or d.k. Hauptschule west Realschule east Abitur + other

72.81%

57.12%

Note: k=33, period of analysis: 2011−12 Note: k=33, period of analysis: 2011−12

Notes: Data from SOEP v29.1. Poverty cutoff k = 33.

48 Figure A.3: Income and Multidimensional Poverty—other years (a) years of analysis 2001–02

50% of y med 60% of y med .25 0.11

0.14 .2 0.07

0.04 0.08 .15

0.06 0.07 0.08 0.05 .1

0.03 0.03 0.04 0.05 0.04 .05 0.03 0.03 0.02 0.02 0 27 33 38 27 33 38 k−cutoff

both−poor IO−poor MDO−poor

period of analysis: 2001−02; y is real net household equivalence income.

(b) years of analysis 2006–07

50% of y med 60% of y med

0.11 .25

0.14 0.07 .2

0.04 0.09 .15 0.06 0.07 0.08 0.06 .1

0.07 0.03 0.03 0.04 0.06 0.04

.05 0.04 0.03 0.02 0 27 33 38 27 33 38 k−cutoff

both−poor IO−poor MDO−poor

period of analysis: 2006−07; y is real net household equivalence income.

Notes: Data from SOEP v29.1. Poverty cutoff k = 33. Threshold for income- poverty is 60% of median income. Underlying income concept is real net household equivalence income.